# Restaurants and catering at 5 locations swapped guesswork for a daily demand forecast

> Restaurant group in Czechia, 4 restaurants plus catering: demand forecast per item, supplier orders at 07:45, waste 7.5% instead of 21%, ≈€10k a month saved.

URL: https://emilus.dev/en/cases/restaurants-catering/
Emilus · hello@emilus.dev · https://t.me/brrr28

1. [Emilus](https://emilus.dev/en/)
2. [Case studies](https://emilus.dev/en/cases/)
3. Restaurants and catering

emilus. Procurement

emilus. Integrator

Status: In production

One kilogram of food in five went in the bin, and the best sellers ran out on Friday evenings. Now the system drafts tomorrow's order for all five locations, the manager approves it with one tap, and by 07:45 the suppliers already have their orders.

Case facts

**Client:**
Restaurant group in Czechia4 restaurants and catering, 6 suppliers

**Service:**
emilus. Procurement · emilus. Integrator · demand forecast and POS data from every location

**Status:**
In production at the clienttomorrow's order every morning for all five locations

**In numbers:**
Food waste: 21% → 7.5%ordering: 40 min → 2 min every morning at each location

The “before” figures, the result and the payback are per client data.

## The client's situation

The group runs four restaurants and a catering business in Czechia and buys food from six suppliers. Each location ordered on its own: every morning a chef or a manager looked at yesterday's sales and wrote the order by eye. Nobody saw the stock of the group as a whole, so the same products were bought twice for neighboring locations, and weekend catering orders piled on top of the restaurant orders.

Nobody counted the waste. When they finally did, one kilogram in five was going in the bin: 21% of all purchases, about €170,000 a year. At the same time, the best sellers kept running out on Friday evenings, the busiest hours of the week.

**21%**

of purchases went to waste: one kilogram in five

*per client data*

**≈€170k**

of food a year thrown away

*per client data*

**≈40 min**

every morning spent on orders at each location

*per client data*

The main rule The model does the math, a human makes the call: orders go to suppliers only after the manager's “ok”.

## What the system does

- **Pulls sales from the POS at all five locations every day,** with no exports or spreadsheets, so the forecast always runs on fresh numbers.
- **Keeps one inventory of products and expiry dates for the whole group:** you see what is stored where, and a product sitting at one location is not bought again for the one next door.
- **Forecasts demand for every item at every location** from 12 months of history, taking in the season, holidays, weather, bookings and prices, so every order follows expected demand.
- **Drafts tomorrow's order for all locations on its own,** with catering on a separate line, so a weekend banquet does not eat into restaurant stock.
- **Marks lines “check” where demand is unusual:** a large catering job, a banquet or a holiday. The manager reviews a few lines instead of the whole list.
- **Shows what changed since yesterday:** weather, a holiday, a new booking. It is clear at a glance why an order is bigger or smaller.
- **Sends orders only after the manager's “ok”:** one tap in Telegram approves all five, and any line can still be changed.
- **Sends the orders to six suppliers at 07:45,** each one gets only its own items, with no morning calls or message threads.
- **Lets the kitchen log waste in 2 minutes per shift,** so you see what goes to waste and at which location, and money no longer disappears unnoticed.
- **Retrains every week on fresh sales and waste data,** so the forecast keeps up with the season and with shifts in demand.

**Illustration** · How the system works

### The architecture in plain terms

POS sales land in the group's shared inventory every day. The model forecasts demand for each item and location, the system drafts tomorrow's orders, the manager approves them in Telegram, and the orders go out to the suppliers.

**Data:**
POS sales by location · stock and expiry dates · kitchen waste log · bookings and catering · holidays and weather · prices

**Roles:**
manager: approval in Telegram · kitchen: waste log · suppliers: orders at 07:45

## How it works

**The daily cycle, from yesterday's sales to supplier orders**

1. location tills · **POS** · Sales from all five locations every day
2. system · **Inventory** · Shared stock and expiry dates across the group
3. model · **Forecast** · Demand per item and location: season, holidays, weather, bookings, prices
4. system · **Order** · For tomorrow, all locations, catering separately
5. manager · Telegram · **Manager's “ok”** · One tap for all five orders
6. system · **Suppliers** · Orders at 07:45, each gets its own
7. kitchen · **Waste** · 2 minutes per shift
8. model · **Retraining** · Every week on fresh sales and waste

Honestly, no forecast will predict a banquet nobody knew about. That is why bookings and catering go straight into the calculation, and the system marks unusual items “check” for the manager.

## Result and payback

**7.5%**

was 21%

of purchases now goes to waste: 2.8 times less

per client data

**≈€10k**

a month no longer ends up in the bin: ≈€120k a year

per client data

**2 min**

was 40 min

every morning at each location: the kitchen logs waste, the system drafts the order itself

per client data

**Project payback: ≈2 months *per client data*** · Waste fell from 21% to 7.5% of purchases. Of the ≈€170k a year that used to go in the bin, ≈€10k a month now stays in the business, and the system paid for itself in full within 2 months.

## What is running

1. running Daily POS sales from all locations and one inventory of products and expiry dates across the group
2. running Forecast per item and location, and tomorrow's order for all five locations, with catering separate
3. running Manager approval in Telegram and orders to the six suppliers at 07:45
4. running Kitchen waste logging in 2 minutes and weekly retraining of the forecast

The service behind this case

## [emilus. Procurement](https://emilus.dev/en/services/procurement/)

Purchase forecasts and ready supplier orders. You get a list of what to order and how much, and you confirm it before it goes out. Price after the brief.

## Other case studies

- [AI front desk for a dental clinic chain](https://emilus.dev/en/cases/dental-clinic-chain/)
- [Demand forecast and a ready order to manufacturers](https://emilus.dev/en/cases/vet-pharma-purchasing/)
- [AI sales assistant in Instagram Direct for a clothing store](https://emilus.dev/en/cases/clothing-store-instagram/)
- [Four AI agents for a car dealer hub](https://emilus.dev/en/cases/car-dealer-hub/)

[All case studies](https://emilus.dev/en/cases/)

Next step

## Write to us. A human replies.

[Message us on Telegram](https://t.me/brrr28)

[All contacts](https://emilus.dev/en/contact/)